What changed

Adds poc-onboarding.mdx at the repo root: a self-serve setup guide for enterprise customers running a Mintlify proof of concept.

Eight steps, from connecting a repository through to a results review, opening with a workflow table so a reader can see the whole shape before starting. The content is derived from the SE working session guides in the internal KB (kb.mintlify.com/sales/se/working-sessions/*), so the customer-facing instructions and the SE-led sessions teach the same mechanics.

Rationale

Sales needs a single link to send enterprise prospects at POC kickoff. The audience is mostly non-technical, so each step names who from the customer's team is needed, for how long, and what to do at the common blockers.

Corrections the working sessions surfaced

• Authentication requires a custom domain or a *.mintlify.app subdomain. It does not work on the default .mintlify.site URL or on a custom basepath. The first draft treated the custom domain as skippable during a POC and auth as independent of it, which would have sent customers into a wall in their final week. Now flagged in step 1 and again in step 7.
• Dashboard access and documentation access are different things — who logs into the dashboard vs who can read the docs. Now split, with the method list (password / Mintlify-managed / OAuth 2.0 / JWT) and the two-account test that actually proves group gating works.

Notable content

• The repo / dashboard / live-site model up front, which non-technical readers need before the steps make sense.
• Step 4 uses the real editor flow, and states that publishing writes a Git commit so CODEOWNERS, required reviews, and branch protection still apply — the objection engineering stakeholders raise first.
• Step 6 asks readers to score assistant answers on whether the right page was cited, not just whether the answer looked right, and points at frontmatter description when a correct page never gets retrieved.
• Sample content is chosen from top pages by traffic rather than intuition, with the hard-to-migrate pages called out explicitly.
• Success is defined as one business goal with a recorded baseline, plus two or three criteria the POC itself can settle.

Hidden-page mechanics

Not listed in docs.json, and carries noindex: true in frontmatter. Per organize/hidden-pages.mdx, omitting a page from docs.json keeps it out of the sidebar, site search, AI assistant context, llms.txt, and the sitemap — but no noindex meta tag is emitted, so a crawler that finds the URL elsewhere can still index it. noindex: true supplies the tag. Same treatment as enterprise-contracting.mdx and migration-services/go-live-checklist.mdx.

Step 7 uses this page itself as the worked example of "hidden is not private."

Why one page and not a series

Considered splitting the steps, or the eleven SE workshops, into separate hidden pages. Tested it first: a hidden: true group in docs.json renders its pages with the visible groups in the sidebar and no next/prev pagination, so a reader lands mid-series looking at the main product docs nav with no way forward. Every link between pages would have to be hand-maintained. Combined with the translation pipeline picking up ./**/*.mdx (three extra locale copies per page), a multi-page series belongs in its own deployment rather than here.

Verification

• mint broken-links and mint a11y — both clean on the final version. Confirmed the link checker does scan pages outside docs.json by temporarily breaking a link and seeing it reported.
• Rendered locally and checked visually: Steps, nested Accordion, AccordionGroup, Tabs, titled JSON code blocks, and the tables.
• All cross-page anchors verified against the target files' headings.
• vale not run — not installed locally.

Areas of uncertainty

1. Step 3 says to ask your Mintlify representative whether migration is included in the POC, rather than promising one. If POC migrations are standard, state it outright.
2. The two-to-three week timeline is an estimate from the step contents, not from real POC data.
3. The Forrester deflection benchmarks are deliberately omitted. The leading-brands-value-from-docs session carries them (~25–30% for a maintained KB, 35–45% with contextual guidance) with a warning that they're directional third-party figures. The page tells customers to baseline their own ticket volume instead. Easy to add back with attribution if preferred.

Also worth confirming the page belongs at the root rather than under migration-services/, where the other two hidden enterprise pages live.

🤖 Generated with Claude Code